RFR: 8155824: JDK build should tune down small apps more aggressively

[Claes Redestad]

On 2016-05-02 16:01, Erik Joelsson wrote:
> Looks good to me.

Thanks!

/Claes

>
> On my machine, the numbers are:
>
> Before:
> real 3m55.455s
> user 48m19.736s
> sys 7m12.815s
>
> With suggested patch:
> real 3m36.529s
> user 41m12.488s
> sys 6m17.110s
>
> /Erik
>
> On 2016-05-02 15:41, Claes Redestad wrote:
>> Hi build-dev!
>>
>> I recently discovered that running with -XX:TieredStopAtLevel=1 can 
>> speed up jmod execution a bit:
>>
>> http://mail.openjdk.java.net/pipermail/jigsaw-dev/2016-May/007593.html
>>
>> Applying this to the runner itself might be a mouthful, but it seems 
>> profitable to apply this to JDK
>> build itself. The question was raised if we would get benefits 
>> elsewhere if we apply it to all "small"
>> java targets...
>>
>> Some experimental results on my machine:
>>
>> time make images
>>
>> real    2m39.281s
>> user    42m15.164s
>> sys    3m56.388s
>>
>> -XX:TieredStopAtLevel=1 for jmod target only:
>>
>> real    2m34.913s
>> user    38m18.112s
>> sys    3m44.528s
>>
>> -XX:TieredStopAtLevel=1 for all small java targets in makefiles:
>>
>> real    2m29.490s
>> user    35m45.944s
>> sys    3m33.428s
>>
>> Reports indicate the real time improvement is even better on smaller 
>> machines.
>>
>> I thus propose this patch:
>> http://cr.openjdk.java.net/~redestad/8155824/webrev.01/
>>
>> Bug: https://bugs.openjdk.java.net/browse/JDK-8155824
>>
>> Thanks!
>>
>> /Claes
>